Tag: hardware
All the articles with the tag "hardware".
-
Codex Micro Is an Agent Control Surface
OpenAI's small Codex keypad is not interesting because it is hardware. It is interesting because agent work is starting to need a dedicated control layer.
-
OpenAI's Jalapeno Chip Is About Owning the Whole Stack
OpenAI's first custom Broadcom inference chip shows frontier AI labs becoming full-stack infrastructure companies, not just model providers.
-
NVIDIA Vera Says Agents Are A CPU Problem Too
NVIDIA's Vera CPU push is a reminder that agentic AI is not only about GPUs. Tool use, sandboxes, memory, and orchestration need serious CPU design.
-
Amazon Bee Shows the Ambient AI Consent Problem
Amazon's Bee wearable points toward proactive ambient AI, but always-available memory has a hard social problem: other people did not opt in.
-
Memory Is Becoming the AI Chip Tax
Epoch AI estimates high-bandwidth memory rose to 63% of AI chip component spending by late 2025. The AI infrastructure story is no longer just about logic dies.
-
Hugging Face Put an App Store on a $299 Robot
Reachy Mini now has an open app store with more than 200 apps. The important part is not the robot, it is the builder loop.
-
The AI Chip Story Just Moved From GPUs to CPUs
Meta signed a massive AWS Graviton deal for agentic AI workloads. GPUs still train the models, but CPUs may run more of the everyday agent work.
-
Eli Lilly Built Pharma's Most Powerful AI Supercomputer. That Is the Entire Point.
LillyPod is 1,016 NVIDIA Blackwell Ultra GPUs on a DGX SuperPOD. Eli Lilly did not hire a consulting firm to write an AI strategy deck. They built a supercomputer. The gap between companies that treat AI as a feature and companies that treat it as infrastructure is about to become permanent.
-
Robotics Startups Raised $1.2 Billion in a Single Week. The Money Has Made Its Decision.
Five robotics companies raised over $1.2 billion in seven days. Mind Robotics, Rhoda AI, Sunday, Oxa, Fauna Robotics. The venture capital world just declared that physical AI is no longer a research bet. It is an investment thesis.
-
Google Just Made Every AI Model 6x Cheaper to Run. Memory Chip Stocks Crashed.
TurboQuant compresses LLM memory from 16 bits to 3 bits with zero accuracy loss. 6x less memory, 8x faster inference. And the stock market panicked, because apparently nobody learned from the DeepSeek episode.